Neurofeedback on Oscillatory Processes Related to Tinnitus.
Introduction
Subjective tinnitus, a condition characterized by the sensation of a sound without any physical source, affects roughly 10% of the general population. In further 10% of the patients, the condition leads to significant decrease of quality of life (Heller, 2003).
Consistent findings over the past years show that a) tinnitus is a disorder of the brain (Eggermont & Roberts, 2004) and b) the underlying cause of tinnitus is a deficit of inhibition triggered by the loss of input to the relevant areas (Weisz, Dohrmann, & Elbert, 2007).
Recent research has led to great insights into the neuronal correlates of tinnitus. A relevant finding is reduced ongoing (spontaneous) alpha power in auditory areas (Lorenz, Müller, Schlee, Hartmann, & Weisz, 2009; Weisz et al., 2005; Weisz, Muller, et al., 2007). Originally attributed to idling of the underlying cortical region (Pfurtscheller et al., 1996), recent research has shown that alpha oscillations rather represent the excitatory-inhibitory balance of underlying cortical areas, with strong alpha representing a state of relative inhibition.
Furthermore, increasing evidence arises that the occurrence of alpha oscillations is not limited to the visual and somatosensory system but is also found in the auditory system with comparable functional correlates (Weisz et al., 2011). On a system level, three main lines of research concerning the power of alpha oscillations exist: 1) active inhibition of cortical areas that would possibly interfere with a current task to be solved (see, e.g., Jensen & Mazaheri, 2010), 2) spontaneous fluctuations of alpha power altering the perception of incoming
61
stimuli or those induced at the cortical level (Min & Herrmann, 2007; Romei, Brodbeck, et al., 2008) and 3) the impact of resting state alpha power on perception (Romei, Rihs, et al., 2008) or its relationship to diseases like tinnitus (Lorenz et al., 2009; Weisz, Dohrmann, et al., 2007; Weisz et al., 2005; Weisz, Muller, et al., 2007). While the association of alpha oscillations and inhibition was based on behavioral associations so far, a recent study by Haegens et al. shows the link between cellular recordings, alpha power and behavioral measurements, strengthening the inhibition hypothesis (Haegens et al., 2011). In conjunction with the aforementioned results on the neural correlates of tinnitus, the hypothesis is that decreased auditory cortical alpha in tinnitus could be a useful proxy for decreased inhibition in the auditory cortex. Interestingly increases of alpha activity after neurofeedback have been reported to lead to significant decreases in distress scores (Crocetti & Forti, 2011; Dohrmann, Elbert, et al., 2007; Dohrmann, Weisz, et al., 2007). Even though the speculation is tempting that these alpha enhancements contributed to a normalization of the disturbed excitatory-inhibitory balance, these previous studies however lack convincing evidence that indeed auditory cortical alpha activity was enhanced.
Moreover, studies exist that used low frequency (putatively inhibitory) repetitive Transcranial Magnetic Stimulation (rTMS) to effectively decrease tinnitus distress, yet neither report changes in ongoing oscillations nor give insight in the exact mechanisms on the cortical level (Folmer et al., 2006; Khedr et al., 2009; Kleinjung et al., 2007; Lorenz, Müller, Schlee, Langguth, & Weisz, 2010). One might speculate that rTMS increases cortical inhibition not only on a short timescale but would shape neuronal networks in a way to establish long-term changes. If this assumption is true, rTMS should lead to an increase of alpha power at respective regions.
62
The distress induced by tinnitus cannot solely be explained by altered activity in lower level auditory areas. It is thus obvious that long-range connectivity between auditory and higher order areas is of high interest as previous studies have already reported corresponding results (Plewnia, 2010; Schlee et al., 2009; Vanneste, Focquaert, & Heyning, 2011). For both, i.e. power and connectivity changes, previous studies have focused on the comparison of groups, thereby not being capable of excluding the possibility that the identified neurophysiological processes are not relevant for tinnitus per se. In the current study, we investigated effects of neuromodulation via neurofeedback and the more established approach of 1 Hz rTMS on resting state alpha power of the auditory cortex and possible links to functional long-range connectivity within tinnitus subjects. Our main intention was to test whether any of these two methods reliably enhances auditory cortical alpha activity and whether these are accompanied by modulations of brain connectivity.
Methods
We report data from two experiments. Subjects took part in either the rTMS or the Neurofeedback study.
Subjects
Initially, 12 patients took part in the Neurofeedback study. One patient decided to stop the treatment because of lack of improvement. Three further patients completed the treatment but had to be excluded from data analysis due to excessive artifacts in the MEG-data (we tolerated a maximum of 6 bad channels and 40% bad trials in the data). This left 8 patients (1 female, mean age ± standard deviation: 57 ± 9) for analysis. On average, these patients had suffered from tinnitus for 5.4 years (standard deviation: 6.4 years). The average distress, assessed with the German version of the Tinnitus Questionnaire (Goebel & Hiller, 1994) was
63
22.6 (standard deviation: 10.0). One patient suffered from mild dysthymia according to the MINI interview (Lecrubier et al., 1997) and was treated with 60mg Amoxid per day. All other patients were free of psychiatric diagnoses and psychoactive medication. The TMS study was conducted on 10 patients of whom one was excluded due to artifacts in the MEG-data. This left 9 patients (2 female, mean age ± standard deviation: 50 ± 15). On average, these patients had suffered from tinnitus for 2.3 years (standard deviation: 1.2 years). The average distress was 26.2 (standard deviation 14.8). All patients in the TMS group were free of psychiatric diagnoses and psychoactive medication. The groups did not differ in age (two-sided T-test, p = 0.25), tinnitus distress (p = 0.9) and tinnitus duration (p = 0.2).
All patients gave written informed consent before participating. The procedures were approved by the Institutional Review Board of the University of Konstanz.
Study Design
Neurofeedback
Patients in the Neurofeedback group received 10 sessions of auditory alpha Neurofeedback over a period of approximately 4 weeks (2-3 sessions per week). Approximately one week before the first session and one week after the last session, 5 minutes of resting state MEG (eyes open) was recorded with a 148-channel whole-head magnetometer system (MAGNES 2500WH, 4d Neuroimaging, San Diego, USA), installed in a magnetically shielded room (Vakuumschmelze Hanau, Germany).
Neurofeedback was conducted using a 32 channel EEG System (Neuroconn, Ilmenau, Germany). The acquired data were processed in real-time with ConSole (Hartmann, Schulz,
& Weisz, 2011) and fed back to the patient via a TFT screen.
64
Each session consisted of one baseline measurement to calibrate the neurofeedback system, four training runs and another baseline measurement after the training to assess changes in cortical activity. In the training runs, patients were shown a feedback on the screen for 5 s without hearing a tone. They were instructed to consider this period as a baseline that showed how auditory areas of their brain behaved without any input. Afterwards, patients were stimulated with a sound that was spectrally filtered to match their tinnitus percept as close as possible for another 5 seconds (Noreña, Micheyl, Chéry-Croze, & Collet, 2002).
Alpha increase above an individually defined threshold for one second within the second 5 s period was rewarded by displaying a smiley on the screen. Thus, we tried to exploit the well know effect that auditory alpha desynchronizes on sensory input (Lehtelä, Salmelin, & Hari, 1997; Mimura et al., 1962; Weisz et al., 2011). The rationale behind this approach was to provide patients with the possible strategy to enhance auditory alpha power by decreasing attention to the sound (Müller & Weisz, 2011). Besides, patients should also be enabled to transfer the strategy of ignoring a “tinnitus-like” sound to ignoring the actual tinnitus percept. Baseline measurements differed from the training runs only by not providing feedback to the patients. The patients were instructed to passively listen to the sounds with eyes open.
Data acquired from 29 electrodes on the scalp and 2 electrodes beside and above the right eye to facilitate artifact correction were sent to ConSole. The DC part of the signal was filtered out using an optimized recursive filter (y(t) = x(t) – x(t-1) + 0.995 * y(t-1)). The data were then lowpass filtered (4th order Butterworth filter; cutoff: 16Hz), average referenced and artifact corrected via ICA (JADE algorithm (Cardoso & Souloumiac, 1993)). The data were then projected onto eight regional sources. The data of the two temporal sources were
65
subjected to frequency analysis and the relative energy at the individual’s alpha frequency was averaged and fed back to a computer screen.
Tinnitus distress was measured at a diagnostic session approximately one week before the first MEG measurement and at the last Neurofeedback session using the German adaption of the tinnitus questionnaire (Goebel & Hiller, 1994).
TMS
Patients in the TMS group received sham and verum treatment in a pseudo-randomized crossover design. To avoid potential carry-over effects, the two stimulation series were separated by three months. Ten sessions of rTMS were conducted on ten consecutive working days using a biphasic MAGSTIM system (Rapid2, MAGSTIM CO., Whitland, Dyfed, UK) and an air-cooled figure-of-eight coil (MAGSTIM Air Film Coil, 70mm). The handle of the coil was pointed upwards. Neuronavigation (Advanced Neuro Technology, Enschede, Netherlands) was used to target the main generator of the auditory N1, contralateral to the predominant tinnitus location (locations derived for all participants from data published by LORENZ). Each rTMS session consisted of 1000 pulses administered at 1 Hz. The intensity was of 50% of maximum stimulator output. For the sham condition the same parameters were applied but the coil was tilted by 45° over one wing. As for the Neurofeedback group, five minutes of resting state MEG were recorded before and after each treatment series. The setup was the same as in the Neurofeedback study. Tinnitus distress was measured using the German version of the TQ (Goebel & Hiller, 1994) when patients came for MEG examination.
Data of the rTMS group with a detailed region of interest analysis are being prepared in a companion article (MÜLLER et al., submitted). The present paper focuses on alpha related changes and modulations of long-range connectivity patterns.
66 Data Analysis
MEG data of both Neurofeedback and rTMS subjects were analyzed using fieldtrip (Oostenveld et al., 2011), an open source toolbox for MEG and EEG analysis in Matlab (The Mathworks). The five minutes resting state data were epoched into segments of two seconds each (no overlap). The resulting epochs were carefully examined for artifacts.
Channels that showed excessive noise or other artifacts below 20Hz were interpolated using spline interpolation (Perrin et al., 1989).
Subsequent analysis was done entirely in source space. We therefore generated equally spaced dipole grids of 5mm and 10mm resolution on the MNI brain provided by the SPM8 toolbox (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/) and morphed the grid to the individual headshapes of the patients. Thus, the individual positions of the grid points in each patient’s brain approximately represented the same anatomical region. Forward models for each patient were computed using the “sensor-weighted overlapping spheres”
algorithm (Huang, Mosher, & Leahy, 1999).
For the power analysis, we first computed the cross spectral density matrix (CSD) for each trial between 8-12Hz (FFT with hanning taper). To calculate the spatial filter for source space projection, the data were first highpass (8Hz, 4th order zero-phase butterworth filter) and lowpass (12Hz, 4th order zero-phase butterworth filter) filtered. The covariance between all channels of the filtered data was used to calculate the spatial filter using the LCMV beamformer algorithm (Veen & Drongelen, 1997) with 15% regularization. The resulting spatial filter was then used to project the CSD matrix to source space using the 5mm grid.
The diagonal of the resulting matrix is the energy of each channel in the respective frequency band as was used for further power analysis.
67
We used the 10mm spaced grid for the connectivity analysis because of memory and computing power constraints. The same forward models as for the power analysis were used. The spatial filter was again confined to activity between 8-12Hz and regularized by 15%. The CSD matrix was then projected to source space. In order to calculate effective connectivity, the phase slope index (PSI) (Nolte & Müller, 2010; Nolte et al., 2008) was calculated between the center frequency ±2Hz. The PSI measures the slope of the difference of the phases of two signals in the frequency domain. As for every measure of effective connectivity, direction is determined by measuring if signal a comes before signal b or vice versa. The PSI exploits the fact that if signal a comes before signal b, the slope of the difference between the phases of the signals is positive while it is negative if signal b comes before signal a. For each subject and each condition (pre and post), a distribution of the resulting PSI values was calculated which was used to threshold the individual connections.
Only connections that showed a PSI value higher or lower than 2 standard deviations were kept. This thresholding provided the adjacency matrix required for computation of node degree (Bullmore & Sporns, 2009), i.e. the sum of each voxel’s connection to other voxels.
Statistical Analysis
In order to assess whether the decrease of tinnitus related impairment differed between three groups (NFB, TMS, Sham), we calculated the relative improvement for each patient in each group ((Pre – Post) / Pre) and analyzed these using a one-factor ANOVA.
Power and node degree of the MEG measurements were compared within each group using a cluster-based non-parametric, permutation-based statistic (Maris & Oostenveld, 2007) that controls the Type I Error with respect to multiple comparisons. First, ordinary t-statistics (post vs. pre, one-sided for power analysis, two-sided for node degree analysis) were
68
calculated. The cluster-finding algorithm identified clusters of neighboring voxels and frequency bins that had p < 0.05. The test statistic for the permutation test was the sum of all t-values in a cluster. The statistic was repeated for shuffled data, in which data were randomly reordered across conditions (null hypothesis stating that power or node degree did not differ between pre and post measurements). Upon each permutation, the cluster with the highest sum of t-values was kept. By these means, a null distribution was created from 1000 permutations and the p-values for the empirically derived clusters could be calculated.
Results
Tinnitus Questionnaires
We compared the effect in all three groups measured with the German version of the TQ.
The ANOVA shows a main effect for treatment group (F = 4.91; p < 0.018). On further investigation the patients in the TMS group did not show a significant improvement for neither treatment. On average, verum treatment led to a small decrease of TQ score from 26.2 (standard deviation: 14.8) to 25.9 (standard deviation: 18.9). Sham treatment decreased the score in the same group from 27.2 (standard deviation: 16.3) to 25.3 (standard deviation: 14.8). In contrast, the Neurofeedback patients had their TQ scores on average decreased from 22.6 (standard deviation: 10.0) to 14.8 (standard deviation: 11.15).
The individual difference is highly significant (p < 0.003) only for the Neurofeedback group according to a paired one-sided t-test (see Fig. 1). Furthermore, the variability of the difference within the sham condition was far higher than in the other two conditions and the variability at the pre and post measurements were far higher for both rTMS conditions compared to the Neurofeedback condition.
69
Figure 1: Average distress decrease for the three conditions involved in the study according the Tinnitus Questionnaire.
The three conditions differ according to an ANOVA (F = 4.91; p < 0.018). Within conditions, only Neurofeedback led to a significant decrease of distress (p < 0.003).
Power Analysis
As stated in the introduction, the main goal of both intervention techniques is the increase of inhibitory activity in auditory cortical regions which should become manifested in increases of alpha oscillations in auditory cortical areas. Yet, we were also interested whether other areas would also be modulated by one of the approaches used. To circumvent the problem of multiple comparisons, we used a cluster-based, non-parametric permutation based approach (Maris & Oostenveld, 2007). However, no significant cluster was found for either condition.
Since our regions of interest were the auditory cortices we also checked the uncorrected results. These results showed a significant increase (p < 0.05) in alpha power in the vicinity of the right auditory cortex after treatment only for the neurofeedback group but not for the TMS group in either (verum or sham) condition (see Fig. 2a). To further investigate whether
70
the three groups differ with respect to alpha power increase at the right auditory cortex, we compared the relative increase at the voxel showing the highest T-value in the Neurofeedback group between all groups using an ANOVA. According to this, the null hypothesis that the average alpha power increase at that specific voxel does not differ between the groups can be rejected (F = 6.94; p < 0.005; Fig. 2b). Neurofeedback is still the only condition in which we saw an alpha power increase after we repeated the analysis using the hemisphere ipsilateral to stimulation for the rTMS conditions. The Neurofeedback result shows strong lateralization, yet fits with our primary hypothesis.
Figure 2: Alpha power increase after treatment at right auditory regions. a) shows voxels in the Neurofeedback condition that showed a significant increase of alpha power. The effect is located at and in the vicinity of BA41. b) Comparison of relative alpha increase for all three conditions. The conditions differ according to an ANOVA (F = 6.94; p < 0.005).
Connectivity Analysis
Recent publications suggest a relationship between local synchronization in the alpha band and long-range connectivity of the specific region to other areas in the brain (Haegens et al., 2011; Jensen & Mazaheri, 2010) even though this idea is still awaiting empirical confirmation. We thus analyzed whether the reported increase of alpha power in the right auditory cortex after neurofeedback training co-occurs with a decrease of that region’s long-range connectivity. The so-called node-degree, a graph-theoretical measure counting the
71
significant connections (here computed via the PSI method) from and to one node (in this case, one voxel) was calculated.
The cluster based strategy for the whole cortex did not reveal any significant cluster in any group or condition. On the uncorrected level, we however found a significant increase of outgoing connections after treatment for the neurofeedback group in a region directly neighboring the alpha power increase (Fig. 3). In order to scrutinize a possible relationship between the two regions, we correlated the voxels showing the highest t-values in each of the regions. This correlation between power and connectivity increase was not significant (r
= 0.27, p = 0.52), which may be due to the low amount of remaining patients. Yet, we found that out of the 8 patients in the analysis 6 showed an increase in power, 7 showed an increase in outgoing connectivity and in 7 patients power and connectivity both either increased or decreased. The null hypothesis of the latter distribution being due to chance can be rejected according to a one-sided binomial test with a significance level of 0.05. This is suggestive of a relationship between alpha power and connectivity changes, albeit not being linear.
72
Figure 3: a) Increase of alpha node degree for outgoing connections after Neurofeedback training. The effect is located at a region neighboring that showing the power increase. As can be seen in subfigure b), the overlap between both regions is very small.
Yet, increases and decreases of long-range connectivity are more ubiquitous than the aforementioned effects for power, i.e. local synchrony. We found decreases of the ingoing node degree at auditory areas for the Neurofeedback group at the left BA41 (Fig. 4). The two rTMS conditions showed an interesting result at right auditory areas more posterior to the Neurofeedback effects: while outgoing connectivity was increased after verum treatment, we found a decrease after sham treatment (Fig. 5).
73
Figure 4: Decrease of alpha node degree for ingoing connections after Neurofeedback training. The effect is located at the left BA41 and neighboring regions.
Figure 5: Effects for outgoing node degree for the sham and TMS conditions. a) After TMS treatment, outgoing node degree was increased at auditory regions. The effect is very similar to the one found after Neurofeedback treatment. Yet, neither behavioral nor power modulation was found. Interestingly, the effect is antithetic to the decrease of outgoing connectivity found in the sham condition (b).
74
Discussion
The relationship between the severity of tinnitus distress and a chronic decrease in alpha synchronization in auditory cortical areas has been proposed in recent years (e.g., Lorenz et al., 2009; Weisz, Dohrmann, et al., 2007; Weisz, Muller, et al., 2007). According to this framework, one therapeutic approach to alleviate tinnitus by normalizing the disturbed
The relationship between the severity of tinnitus distress and a chronic decrease in alpha synchronization in auditory cortical areas has been proposed in recent years (e.g., Lorenz et al., 2009; Weisz, Dohrmann, et al., 2007; Weisz, Muller, et al., 2007). According to this framework, one therapeutic approach to alleviate tinnitus by normalizing the disturbed